[Paper Review] Optimal Data Detection in Large MIMO

Charles Jeon, Ramina Ghods+2 more|arXiv (Cornell University)|Nov 5, 2018

Advanced MIMO Systems OptimizationEngineering71 references19 citations

TL;DR

This paper proposes LAMA (Large MIMO Approximate Message Passing), a computationally efficient data detection algorithm for large MIMO systems that achieves the same error-rate performance as the individually optimal (IO) detector under i.i.d. Rayleigh fading and specific constellation assumptions. By leveraging complex-valued Bayesian approximate message passing with state evolution, LAMA asymptotically decouples the MIMO channel into independent AWGN channels with equal SNR, enabling precise performance and complexity analysis in the large-system limit, with strong finite-dimensional performance close to optimal at low complexity.

ABSTRACT

Large multiple-input multiple-output (MIMO) appears in massive multi-user MIMO and randomly-spread code-division multiple access (CDMA)-based wireless systems. In order to cope with the excessively high complexity of optimal data detection in such systems, a variety of efficient yet sub-optimal algorithms have been proposed in the past. In this paper, we propose a data detection algorithm that is computationally efficient and optimal in a sense that it is able to achieve the same error-rate performance as the individually optimal (IO) data detector under certain assumptions on the MIMO system matrix and constellation alphabet. Our algorithm, which we refer to as LAMA (short for large MIMO AMP), builds on complex-valued Bayesian approximate message passing (AMP), which enables an exact analytical characterization of the performance and complexity in the large-system limit via the state-evolution framework. We derive optimality conditions for LAMA and investigate performance/complexity trade-offs. As a byproduct of our analysis, we recover classical results of IO data detection for randomly-spread CDMA. We furthermore provide practical ways for LAMA to approach the theoretical performance limits in realistic, finite-dimensional systems at low computational complexity.

Motivation & Objective

• To develop a low-complexity data detection algorithm for large MIMO systems that matches the performance of the individually optimal (IO) detector.
• To enable exact analytical characterization of performance and complexity in the large-system limit using the state-evolution framework.
• To investigate performance-complexity trade-offs in practical, finite-dimensional MIMO systems.
• To recover classical IO detection results for randomly-spread CDMA as a byproduct of the analysis.

Proposed method

• Proposes LAMA, a complex-valued Bayesian approximate message passing (AMP) algorithm tailored for large MIMO systems.
• Employs the state-evolution framework to analytically track the noise variance σ²ₜ at each iteration t, enabling precise performance and complexity analysis.
• Derives optimality conditions under which LAMA achieves the same error-rate performance as the IO detector, assuming i.i.d. Rayleigh fading and specific constellation sets (e.g., QPSK, PAM, PSK).
• Uses the decoupling property of AMP to model the large MIMO system as parallel, independent AWGN channels with equal SNR in the asymptotic regime.
• Introduces a separable constellation model to enable analytical tractability of the message-passing updates and MSE functions.
• Derives critical thresholds (e.g., β_min^O, β_max^O) that determine convergence behavior and performance limits via fixed-point analysis of the MSE function.

Experimental results

Research questions

• RQ1 Under what conditions does LAMA achieve the same error-rate performance as the individually optimal (IO) data detector in large MIMO systems?
• RQ2 How does the performance of LAMA scale with system dimensions in the large-system limit (MT → ∞, β = MT/MR)?
• RQ3 What is the precise performance-complexity trade-off of LAMA, and how can it be analytically characterized without extensive simulations?
• RQ4 Can LAMA recover classical results of IO detection for randomly-spread CDMA systems through its theoretical framework?
• RQ5 How close can LAMA get to the theoretical IO performance limit in finite-dimensional MIMO systems with low computational complexity?

Key findings

• LAMA achieves the same symbol error rate (SER) as the individually optimal (IO) detector in the large-system limit (MT → ∞, β = 1) for i.i.d. Rayleigh fading and standard constellations like QPSK.
• In finite-dimensional systems (e.g., 128×128 MIMO), LAMA closely approaches the theoretical IO performance limit, with SER within 1 dB of the optimal AWGN performance at SER = 10⁻³.
• The state-evolution framework enables exact analytical characterization of LAMA’s performance and complexity, allowing precise prediction of required iterations to approach the theoretical limit.
• LAMA converges exponentially fast to its fixed-point solution when β < β_min^O, where β_min^O is the minimum SNR threshold derived from the MSE function’s derivative.
• For separable constellations (e.g., QAM), the complex-valued MSE function Ψ(σ², σ²) is shown to be equivalent to twice the real-valued MSE function Ψ_R(σ²/2, σ²/2), enabling analytical tractability.
• The paper recovers classical IO detection results for randomly-spread CDMA as a special case of the LAMA framework, validating its generality and correctness.